Unmanned aerial vehicles (UAVs) are increasingly used for time-critical disaster response tasks such as search, rescue, and situational assessment. However, coordinating heterogeneous UAV swarms remains challenging due to dynamic task demands, limited onboard resources, intermittent communication, noisy sensing, and platform failures. This paper presents a resource-aware coalition-based task assignment and path-planning (RCTP) framework that integrates model-agnostic meta-learning (MAML) with multi-agent deep deterministic policy gradient (MA-DDPG) to enable rapid adaptation and scalable decentralized coordination in realistic disaster environments. RCTP introduces an end-to-end learning-based coordination framework for multi-UAV disaster response, where MAML provides meta-initialized policies that adapt to new disaster scenarios with only a few gradient updates, while MA-DDPG supports continuous-action control for coalition formation, task assignment, and energy-aware path planning. The framework explicitly models UAV heterogeneity, intermittent LoS/NLoS communication, dynamic obstacles, noisy observations, and UAV failures. Coalition formation is guided by interpretable and resource-aware suitability scores that are updated online, enabling decentralized task reassignment, fault tolerance, and load balancing without relying on centralized infrastructure. Extensive simulations with 10-30 UAVs across heterogeneous tasks show that RCTP outperforms the existing schemes (i.e., PPO, DQN, and MA-DDPG), achieving 30-40% faster mission completion and 10-20% lower energy consumption, higher task success rates, and improved robustness under multiple failures. Complexity analysis confirms lightweight online execution with millisecond-level per-agent inference, demonstrating scalability and real-time feasibility for disaster response. These results establish RCTP as a practical, robust, and adaptive solution for multi-UAV disaster response.
Poudel et al. (2026) studied this question.